6th Plenary Meeting Report
PVEOut project
6th Plenary Meeting
Report
(Caen, 20-21 January 2003)
Participants to the meeting:
- Alan Colchester (UKENT - Great Britain)
- Ali Hojjat (UKENT - Great Britain)
- Judith Sovago (KI – Sweden)
- Bruno Alfano (CNR – Italy)
- Claus Svarer (RHC – Denmark)
- Karim Berkouk (Inserm – France)
- Laszlo Balkay (DPC – Hungary)
- Mario Quarantelli (CNR – Italy)
- Miklos Emri (DPC – Hungary)
- Olaf B. Paulson (RHC – Denmark)
- Tim Dyrby (RHC – Denmark)
Attachments:
Attachment 1: Presentation by Karim Berkouk on Preliminary Results of a voxel-based comparison of GM volume and Glucose uptake in Alzheimer's Disease vs. NV
Attachment 2: Presentation by Karim Berkouk on Preliminary Results of PVC in Alzheimer's Disease
Attachment 3: Presentation by Karim Berkouk on Preliminary Results of PVC using STEPBrain studies
Attachment 4: Presentation by Miklos Emri on Possible Alternative PVEOut development platforms
Attachment 5: “A way to realize the PVELab”
Introduction to the ideas behind the PVElab and going through
"A way to realise the PVElab"
(Tim Dyrby)
An extensive discussion on the implementation of PVEOut software within a single main program called “PVELab” was guided by Tim Dyrby, who posed some basic questions to the consortium that needed an answer before proceeding with the actual PVELab implementation.
Those questions are reported with the answers that were given by the participants.
●Does PVELab cover our need?
After analysing the first It was clear that the first draft of PVELab fitted with the general description as emerged fro the project: a program written in Matlab capable of handling all the PVEOut modules, constituting a user-friendly GUI and monitoring all the processes which are run during a PVE correction process
●Is PVELab flexible enough?
It was remarked that the present PVELab set-up was extremely flexible, and provided a simple tool to manage
●Alternative proposals: GUI or processing- flow?
●How to split the implementation tasks between the partners?
After an extensive discussion It was agreed that each partner in charge of a module was going to write the software for integrating it correctly in PVELab, while
●Timetable?
The proposed timetable is as follow:
Software design (20/1-1/3)
SW Main GUI design (1/3-1/5)
SW Methods (alignment, ROI, PVE correct) design (1/3-1/6)
Main GUI implementation (1/5-1/8)
Methods implementation (1/6-1/10)
SW test and integration (1/10-03 - 1/1-04)
System test (1/1-04 - 1/3-04)
The timetable proposed by RHC was accepted as temporary, pending the EC decision on project extension.
PVC on Alzheimer
(Karim Berkouk)
Dr. Berkouk presented the results of the analysis of the FDG PET and MRI studies from 23 Alzheimer's patients versus 13 normal volunteers using the first prototype of the PVE correction module. The presentation included the descriptionof the modification of the Talairach-based ROI module (published in NeuroImage 2002;17:373-384), to allow the use of the normalization matrix obtained when using SPM.
Left parietal and right posterior cingulate resulted the only regions with a decrease in GMR exceeding the degree of atrophy.
In the discussion it was pointed out that the basal ganglia ROI remained less accurate as compared to the cortical regions, confirming the need to include in PVELab, as defined in the original project, a tool for manual editing of the labelling of these structures, which should be performed when they are a major focus of the studies being analysed.
Dr. Quarantelli confirmed that a ROI editing tool was going to be implemented in PVELab at Naples.
PVE correction Validation and Error Assessment
(Mario Quarantelli)
Dr. Quarantelli presented the final results of the validation and comparison of the PVE techniques on a simulated PET, performed according to what had been decided at the Copenhagen plenary meeting.
To evaluate the four PVEc techniques, the software was applied to a simulated FDG-PET study subsequently introducing larger experimental errors, including coregistration errors (0 to 6 pixel misregistration), segmentation errors (-9.7% to +10,1% GM volume change) and resolution estimate errors (-16.9% to 26.8% FWHM mismatch).
Even in absence of segmentation and coregistration errors, the uncorrected PET values showed -32.1% GM underestimation and 96.6% WM overestimation.
MPVEc left a residual underestimation of GM values (-20.4%). Application of RPVEc and mMGPVEc provided an accuracy above 97%.
The coefficient of variation of the GM ROIs, a measure of the imprecision of the GM concentration estimates, was 6.5% for uncorrected PET data, and increased with PVEc, reaching 7.9% for RPVEc. Co-registration errors appeared to be the major determinant of the imprecision.
Coupling of automated ROI placement and PVEc provides a tool for integrated analysis of brain PET/MRI data, which allows a recovery of true GM ROI values, with a high degree of accuracy when RPVEc or mMGPVEc are used, and a concurrent limited increase of noise in the data.
Among the four tested PVEc methods, RPVEc showed the greatest accuracy, and is suitable when corrected images are not specifically needed. Otherwise, if corrected images are desired, the mMG method appears the most adequate, showing a similar accuracy.
The had been carried out mainly by the CNR, Inserm, RHC and UDEB-PETC centers, and it was decided to submit the manuscript to the Journal of Nuclear Medicine to guarantee the highest possible diffusion (in the field of the Nuclear Medicine, JNM is the most diffuse journal and has the highest impact factor).
GM heterogeneity
(Karim Berkouk)
Karim Berkouk presented a study aiming at comparing, at the same resolution, PET images and the underlying cerebral structure (i.e. MRI) in terms of GM and WM content. The method was first applied to normal volunteers (NV) and Alzheimer's patient studies, and then a conjunction analysis was carried out between these mismatches in NV and AD.
First, the method consists of correcting the mean white and gray matter tracer uptakes from PVE in the Raw PET data. These PVE-corrected GM and WM values, along with the MRI of the same patient, are used to generate a ‘Virtual PET’ image. Accordingly we can compare the two sets of Raw PET and Virtual PET images, whose activity is artificially made proportional to its underlying gray matter volume (GMV). We used two types of comparison: a voxel-based statistical analysis (i.e. SPM) that exhibits clusters with a significant mismatch between PET activity and GM volume, and a region of interest (ROI) analysis that makes the quantification of the mismatch possible.
The method was applied to resting-state FDG-PET data set of normal subjects. We found a relative hyperactivity, as compared to their GM volume, of the cortical and subcortical GM structures (except the thalamus) of the dorsal part of the brain and a relative hypo-activity in infratentorial structures (mainly the hippocampus and cerebellum) and in the mesial temporal cortex, showing that the heterogeneity in FDG distribution observed in PET images is not only due to local differences in grey matter volumes.
The subsequent conjunction analysis revealed a cluster in the posterior cingulate of AD patients which showed a lower metabolic activity as compared to what would be expected based on the atrophy.
Phantom (MRI and PET protocol)
(M. Quarantelli and K. Berkouk)
Karim Berkouk presented the first images of PET and MR scans of the phantom obtained at Inserm.
It was decided to have the current version of the phantom go firs to Copenhagen and then to Stockholm to acquire other PET images with different resolutions
Due to an error in the previous description of the filling procedure by CNR, GM and WM solutions were inverted. As a result, the T2w images had the T1w contrast, and were used for to set up a PVE correction procedure. The following concentrations and acquisition protocols should be used:
MRI protocol:
GM: 100ml of Iron Nanocolloid (Endorem, Guerbet) in 1 l of distilled water
WM: 25mg of MnCl in 1l (198.6mmol/l)
2200/15-90 msec (TR/TE1-TE2 Conventional Spin-Echo images for Proton Density/T2-weighetd images; 4mm-thick contiguous slices covering the whole brain)
600/15 msec (TR/TE Conventional Spin-Echo images for T1w images4mm-thick contiguous slices covering the whole brain )
3D magnetization-prepared spoiled Gradient-echo T1w volume (1mm-thick contiguous partitions covering the whole brain)
PET protocol:
WM: 0.173mCi in 400ml of distilled water
GM: 1.18 mCi in 800 ml of distilled water
both 2d and 3d acquisition including transmission/emission acquisition for attenuation correction, exactly as performed for routine brain FDG studies.
In the current phantom configuration, the presence of large air bubles due to an excessive length of filling tubes which protrude within the phantom precluding a complete filling, resulting in the PETstudy in air in the GM of the frontal pole.
After these tests, and based on the feed-back from the groups scaning the phantom, CNR will send to the consortium a description of the filling procedure.
Prosecution of the project
(Bruno Alfano)
According to the original technical annex, the software modules have been designed and produced so that each module can be used both separately (i.e. each module can be runned by command line) as well as handled from a common main program.
This last option was meant to provide a mor user-friendly package, suitable for wider distribution.
During the second year of project, there has been a delay in obtaining the first prototype of the common handling program, a part of the PVEOut software responsibility of Partner N° 7 of the consortium (Rasna Imaging Systems.
In particular Partner 7 was supposed to provide a common Graphical User Interface (GUI) to allow an easy use of the modules developped by the other partners.
Despite the lack of such a tool, the consortium has been able to achieve in due time all the deliverables and milestones which were originally planned for the first two years of project.
In this respect it is of note that the results obtained so far have fullfilled the expectations, as commented also by the mid-term reviewer ( "... this project is – in total – scientifically interesting, running well, on target, time and budget, and well managed...")
For the last part of the project, however, a GUI and a common main program are necessary, as the consortium will need to apply the software to larger data sets, making the use of separate modules time-consuming and prone to errors.
Furthermore, there is a need to distribute the software to external (i.e. not directly involved in PVEOut) research centers (a strategy also suggested also by the mid-term reviewer) to increase the initial core of users thus increasing the industrial interest of the PVEOut results.
These tasks are nearly impossible without the integration of all the PVEOut modules under a single "main" program.
It was thus confirmed the decision of all the partners to request Rasna to comply to their contractual obligations, and in case of persistent lack of collaboration by Rasna, to terminate their participation to the project according to Article 7.3.b of the Annex II of the contract.
Furthermore, it was also decided to apply for a extension (12 months) of the time frame of the project.
Finally, It was decided that the dissemination to other research groups of the results of the project, within the time frame of PVEOut, needed to obtain a feedback from qualified potential end-users, was going to be scheduled once the first version of the main program was available.
Responsible persons of each centre were requested to send a list of three groups to the coordinator.
Issues for the application to actual PET studies
(Judit Sovago)
It was discussed the analysis that Judit Sovago is going to carry out at KI on Neuroreceptor studies in normal aging and Huntington patients.
Problems for the analysis include, beside the need to refine the Basal Ganglia ROIs (see discussion to Berkouk presentation on results in Alzheimer), the need to run the software on each frame of dynamic scans in sequence, and possibly the importation of ROI sets fromKI software.
CNR is going to modify the PVE correction module to allow for this type of studies and arrangments were made to synchronize this activity with KI analysis, which will constitute a “template” for the design and implementation of these software modifications.
Possible other PVEOut plaforms
(Miklos Emri)
Miklos presented un update on the work carried out in Debrecen, under different R&D projects, which merged in BrainCAD, in the field of medical image post-processing, that has provided potential tools for additional development within PVEOut.
These proramming tools would allow to develop a version of PVELab entirely in C++.
From the discussion emerged that it was felt it is too soon to to drop the Matlab environment, especially considering that henew Matlab version provides a Matlab Virtual Machine that would allow to distribute the PVELab without even needing a Matlab license.
However, in view of the excellent flexibility of the development tools and considering the appeal that a C-version of the software may have for the industry, it was decided that UDEB further developped in C the parts of the software that are not already in that language, integrating a C++version of PVEOut modules in the BrainCAD project.
Attachment 1
Presentation by Karim Berkuk on Preliminary Results of a voxel-based comparison of GM volume and Glucose uptake in Alzheimer's Disease vs. NV
Attachment 2
Presentation by Karim Berkuk on Preliminary Results of PVC in Alzheimer's Disease
Attachment 3
Presentation by Karim Berkuk on Preliminary Results of PVC using STEPBain studies
Attachment 4
Presentation by Miklos Emri on Possible Alternative PVEOut development platforms
Attachment 5
A way to realise the PVELab
by Tim Dyrby